fs/notify/fanotify/fanotify_user.c

   1 #include <linux/fanotify.h>
   2 #include <linux/fcntl.h>
   3 #include <linux/file.h>
   4 #include <linux/fs.h>
   5 #include <linux/anon_inodes.h>
   6 #include <linux/fsnotify_backend.h>
   7 #include <linux/init.h>
   8 #include <linux/mount.h>
   9 #include <linux/namei.h>
  10 #include <linux/poll.h>
  11 #include <linux/security.h>
  12 #include <linux/syscalls.h>
  13 #include <linux/slab.h>
  14 #include <linux/types.h>
  15 #include <linux/uaccess.h>
  16 #include <linux/compat.h>
  17
  18 #include <asm/ioctls.h>
  19
  20 #include "../../mount.h"
  21 #include "../fdinfo.h"
  22 #include "fanotify.h"
  23
  24 #define FANOTIFY_DEFAULT_MAX_EVENTS     16384
  25 #define FANOTIFY_DEFAULT_MAX_MARKS      8192
  26 #define FANOTIFY_DEFAULT_MAX_LISTENERS  128
  27
  28 /*
  29  * All flags that may be specified in parameter event_f_flags of fanotify_init.
  30  *
  31  * Internal and external open flags are stored together in field f_flags of
  32  * struct file. Only external open flags shall be allowed in event_f_flags.
  33  * Internal flags like FMODE_NONOTIFY, FMODE_EXEC, FMODE_NOCMTIME shall be
  34  * excluded.
  35  */
  36 #define FANOTIFY_INIT_ALL_EVENT_F_BITS                          ( \
  37                 O_ACCMODE       | O_APPEND      | O_NONBLOCK    | \
  38                 __O_SYNC        | O_DSYNC       | O_CLOEXEC     | \
  39                 O_LARGEFILE     | O_NOATIME     )
  40
  41 extern const struct fsnotify_ops fanotify_fsnotify_ops;
  42
  43 static struct kmem_cache *fanotify_mark_cache __read_mostly;
  44 struct kmem_cache *fanotify_event_cachep __read_mostly;
  45 struct kmem_cache *fanotify_perm_event_cachep __read_mostly;
  46
  47 /*
  48  * Get an fsnotify notification event if one exists and is small
  49  * enough to fit in "count". Return an error pointer if the count
  50  * is not large enough.
  51  *
  52  * Called with the group->notification_mutex held.
  53  */
  54 static struct fsnotify_event *get_one_event(struct fsnotify_group *group,
  55                                             size_t count)
  56 {
  57         BUG_ON(!mutex_is_locked(&group->notification_mutex));
  58
  59         pr_debug("%s: group=%p count=%zd\n", __func__, group, count);
  60
  61         if (fsnotify_notify_queue_is_empty(group))
  62                 return NULL;
  63
  64         if (FAN_EVENT_METADATA_LEN > count)
  65                 return ERR_PTR(-EINVAL);
  66
  67         /* held the notification_mutex the whole time, so this is the
  68          * same event we peeked above */
  69         return fsnotify_remove_first_event(group);
  70 }
  71
  72 static int create_fd(struct fsnotify_group *group,
  73                      struct fanotify_event_info *event,
  74                      struct file **file)
  75 {
  76         int client_fd;
  77         struct file *new_file;
  78
  79         pr_debug("%s: group=%p event=%p\n", __func__, group, event);
  80
  81         client_fd = get_unused_fd_flags(group->fanotify_data.f_flags);
  82         if (client_fd < 0)
  83                 return client_fd;
  84
  85         /*
  86          * we need a new file handle for the userspace program so it can read even if it was
  87          * originally opened O_WRONLY.
  88          */
  89         /* it's possible this event was an overflow event.  in that case dentry and mnt
  90          * are NULL;  That's fine, just don't call dentry open */
  91         if (event->path.dentry && event->path.mnt)
  92                 new_file = dentry_open(&event->path,
  93                                        group->fanotify_data.f_flags | FMODE_NONOTIFY,
  94                                        current_cred());
  95         else
  96                 new_file = ERR_PTR(-EOVERFLOW);
  97         if (IS_ERR(new_file)) {
  98                 /*
  99                  * we still send an event even if we can't open the file.  this
 100                  * can happen when say tasks are gone and we try to open their
 101                  * /proc files or we try to open a WRONLY file like in sysfs
 102                  * we just send the errno to userspace since there isn't much
 103                  * else we can do.
 104                  */
 105                 put_unused_fd(client_fd);
 106                 client_fd = PTR_ERR(new_file);
 107         } else {
 108                 *file = new_file;
 109         }
 110
 111         return client_fd;
 112 }
 113
 114 static int fill_event_metadata(struct fsnotify_group *group,
 115                                struct fanotify_event_metadata *metadata,
 116                                struct fsnotify_event *fsn_event,
 117                                struct file **file)
 118 {
 119         int ret = 0;
 120         struct fanotify_event_info *event;
 121
 122         pr_debug("%s: group=%p metadata=%p event=%p\n", __func__,
 123                  group, metadata, fsn_event);
 124
 125         *file = NULL;
 126         event = container_of(fsn_event, struct fanotify_event_info, fse);
 127         metadata->event_len = FAN_EVENT_METADATA_LEN;
 128         metadata->metadata_len = FAN_EVENT_METADATA_LEN;
 129         metadata->vers = FANOTIFY_METADATA_VERSION;
 130         metadata->reserved = 0;
 131         metadata->mask = fsn_event->mask & FAN_ALL_OUTGOING_EVENTS;
 132         metadata->pid = pid_vnr(event->tgid);
 133         if (unlikely(fsn_event->mask & FAN_Q_OVERFLOW))
 134                 metadata->fd = FAN_NOFD;
 135         else {
 136                 metadata->fd = create_fd(group, event, file);
 137                 if (metadata->fd < 0)
 138                         ret = metadata->fd;
 139         }
 140
 141         return ret;
 142 }
 143
 144 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 145 static struct fanotify_perm_event_info *dequeue_event(
 146                                 struct fsnotify_group *group, int fd)
 147 {
 148         struct fanotify_perm_event_info *event, *return_e = NULL;
 149
 150         spin_lock(&group->fanotify_data.access_lock);
 151         list_for_each_entry(event, &group->fanotify_data.access_list,
 152                             fae.fse.list) {
 153                 if (event->fd != fd)
 154                         continue;
 155
 156                 list_del_init(&event->fae.fse.list);
 157                 return_e = event;
 158                 break;
 159         }
 160         spin_unlock(&group->fanotify_data.access_lock);
 161
 162         pr_debug("%s: found return_re=%p\n", __func__, return_e);
 163
 164         return return_e;
 165 }
 166
 167 static int process_access_response(struct fsnotify_group *group,
 168                                    struct fanotify_response *response_struct)
 169 {
 170         struct fanotify_perm_event_info *event;
 171         int fd = response_struct->fd;
 172         int response = response_struct->response;
 173
 174         pr_debug("%s: group=%p fd=%d response=%d\n", __func__, group,
 175                  fd, response);
 176         /*
 177          * make sure the response is valid, if invalid we do nothing and either
 178          * userspace can send a valid response or we will clean it up after the
 179          * timeout
 180          */
 181         switch (response) {
 182         case FAN_ALLOW:
 183         case FAN_DENY:
 184                 break;
 185         default:
 186                 return -EINVAL;
 187         }
 188
 189         if (fd < 0)
 190                 return -EINVAL;
 191
 192         event = dequeue_event(group, fd);
 193         if (!event)
 194                 return -ENOENT;
 195
 196         event->response = response;
 197         wake_up(&group->fanotify_data.access_waitq);
 198
 199         return 0;
 200 }
 201 #endif
 202
 203 static ssize_t copy_event_to_user(struct fsnotify_group *group,
 204                                   struct fsnotify_event *event,
 205                                   char __user *buf)
 206 {
 207         struct fanotify_event_metadata fanotify_event_metadata;
 208         struct file *f;
 209         int fd, ret;
 210
 211         pr_debug("%s: group=%p event=%p\n", __func__, group, event);
 212
 213         ret = fill_event_metadata(group, &fanotify_event_metadata, event, &f);
 214         if (ret < 0)
 215                 return ret;
 216
 217         fd = fanotify_event_metadata.fd;
 218         ret = -EFAULT;
 219         if (copy_to_user(buf, &fanotify_event_metadata,
 220                          fanotify_event_metadata.event_len))
 221                 goto out_close_fd;
 222
 223 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 224         if (event->mask & FAN_ALL_PERM_EVENTS)
 225                 FANOTIFY_PE(event)->fd = fd;
 226 #endif
 227
 228         if (fd != FAN_NOFD)
 229                 fd_install(fd, f);
 230         return fanotify_event_metadata.event_len;
 231
 232 out_close_fd:
 233         if (fd != FAN_NOFD) {
 234                 put_unused_fd(fd);
 235                 fput(f);
 236         }
 237         return ret;
 238 }
 239
 240 /* intofiy userspace file descriptor functions */
 241 static unsigned int fanotify_poll(struct file *file, poll_table *wait)
 242 {
 243         struct fsnotify_group *group = file->private_data;
 244         int ret = 0;
 245
 246         poll_wait(file, &group->notification_waitq, wait);
 247         mutex_lock(&group->notification_mutex);
 248         if (!fsnotify_notify_queue_is_empty(group))
 249                 ret = POLLIN | POLLRDNORM;
 250         mutex_unlock(&group->notification_mutex);
 251
 252         return ret;
 253 }
 254
 255 static ssize_t fanotify_read(struct file *file, char __user *buf,
 256                              size_t count, loff_t *pos)
 257 {
 258         struct fsnotify_group *group;
 259         struct fsnotify_event *kevent;
 260         char __user *start;
 261         int ret;
 262         DEFINE_WAIT_FUNC(wait, woken_wake_function);
 263
 264         start = buf;
 265         group = file->private_data;
 266
 267         pr_debug("%s: group=%p\n", __func__, group);
 268
 269         add_wait_queue(&group->notification_waitq, &wait);
 270         while (1) {
 271                 mutex_lock(&group->notification_mutex);
 272                 kevent = get_one_event(group, count);
 273                 mutex_unlock(&group->notification_mutex);
 274
 275                 if (IS_ERR(kevent)) {
 276                         ret = PTR_ERR(kevent);
 277                         break;
 278                 }
 279
 280                 if (!kevent) {
 281                         ret = -EAGAIN;
 282                         if (file->f_flags & O_NONBLOCK)
 283                                 break;
 284
 285                         ret = -ERESTARTSYS;
 286                         if (signal_pending(current))
 287                                 break;
 288
 289                         if (start != buf)
 290                                 break;
 291
 292                         wait_woken(&wait, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
 293                         continue;
 294                 }
 295
 296                 ret = copy_event_to_user(group, kevent, buf);
 297                 /*
 298                  * Permission events get queued to wait for response.  Other
 299                  * events can be destroyed now.
 300                  */
 301                 if (!(kevent->mask & FAN_ALL_PERM_EVENTS)) {
 302                         fsnotify_destroy_event(group, kevent);
 303                         if (ret < 0)
 304                                 break;
 305                 } else {
 306 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 307                         if (ret < 0) {
 308                                 FANOTIFY_PE(kevent)->response = FAN_DENY;
 309                                 wake_up(&group->fanotify_data.access_waitq);
 310                                 break;
 311                         }
 312                         spin_lock(&group->fanotify_data.access_lock);
 313                         list_add_tail(&kevent->list,
 314                                       &group->fanotify_data.access_list);
 315                         spin_unlock(&group->fanotify_data.access_lock);
 316 #endif
 317                 }
 318                 buf += ret;
 319                 count -= ret;
 320         }
 321         remove_wait_queue(&group->notification_waitq, &wait);
 322
 323         if (start != buf && ret != -EFAULT)
 324                 ret = buf - start;
 325         return ret;
 326 }
 327
 328 static ssize_t fanotify_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
 329 {
 330 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 331         struct fanotify_response response = { .fd = -1, .response = -1 };
 332         struct fsnotify_group *group;
 333         int ret;
 334
 335         group = file->private_data;
 336
 337         if (count > sizeof(response))
 338                 count = sizeof(response);
 339
 340         pr_debug("%s: group=%p count=%zu\n", __func__, group, count);
 341
 342         if (copy_from_user(&response, buf, count))
 343                 return -EFAULT;
 344
 345         ret = process_access_response(group, &response);
 346         if (ret < 0)
 347                 count = ret;
 348
 349         return count;
 350 #else
 351         return -EINVAL;
 352 #endif
 353 }
 354
 355 static int fanotify_release(struct inode *ignored, struct file *file)
 356 {
 357         struct fsnotify_group *group = file->private_data;
 358
 359 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 360         struct fanotify_perm_event_info *event, *next;
 361         struct fsnotify_event *fsn_event;
 362
 363         /*
 364          * Stop new events from arriving in the notification queue. since
 365          * userspace cannot use fanotify fd anymore, no event can enter or
 366          * leave access_list by now either.
 367          */
 368         fsnotify_group_stop_queueing(group);
 369
 370         /*
 371          * Process all permission events on access_list and notification queue
 372          * and simulate reply from userspace.
 373          */
 374         spin_lock(&group->fanotify_data.access_lock);
 375         list_for_each_entry_safe(event, next, &group->fanotify_data.access_list,
 376                                  fae.fse.list) {
 377                 pr_debug("%s: found group=%p event=%p\n", __func__, group,
 378                          event);
 379
 380                 list_del_init(&event->fae.fse.list);
 381                 event->response = FAN_ALLOW;
 382         }
 383         spin_unlock(&group->fanotify_data.access_lock);
 384
 385         /*
 386          * Destroy all non-permission events. For permission events just
 387          * dequeue them and set the response. They will be freed once the
 388          * response is consumed and fanotify_get_response() returns.
 389          */
 390         mutex_lock(&group->notification_mutex);
 391         while (!fsnotify_notify_queue_is_empty(group)) {
 392                 fsn_event = fsnotify_remove_first_event(group);
 393                 if (!(fsn_event->mask & FAN_ALL_PERM_EVENTS))
 394                         fsnotify_destroy_event(group, fsn_event);
 395                 else
 396                         FANOTIFY_PE(fsn_event)->response = FAN_ALLOW;
 397         }
 398         mutex_unlock(&group->notification_mutex);
 399
 400         /* Response for all permission events it set, wakeup waiters */
 401         wake_up(&group->fanotify_data.access_waitq);
 402 #endif
 403
 404         /* matches the fanotify_init->fsnotify_alloc_group */
 405         fsnotify_destroy_group(group);
 406
 407         return 0;
 408 }
 409
 410 static long fanotify_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 411 {
 412         struct fsnotify_group *group;
 413         struct fsnotify_event *fsn_event;
 414         void __user *p;
 415         int ret = -ENOTTY;
 416         size_t send_len = 0;
 417
 418         group = file->private_data;
 419
 420         p = (void __user *) arg;
 421
 422         switch (cmd) {
 423         case FIONREAD:
 424                 mutex_lock(&group->notification_mutex);
 425                 list_for_each_entry(fsn_event, &group->notification_list, list)
 426                         send_len += FAN_EVENT_METADATA_LEN;
 427                 mutex_unlock(&group->notification_mutex);
 428                 ret = put_user(send_len, (int __user *) p);
 429                 break;
 430         }
 431
 432         return ret;
 433 }
 434
 435 static const struct file_operations fanotify_fops = {
 436         .show_fdinfo    = fanotify_show_fdinfo,
 437         .poll           = fanotify_poll,
 438         .read           = fanotify_read,
 439         .write          = fanotify_write,
 440         .fasync         = NULL,
 441         .release        = fanotify_release,
 442         .unlocked_ioctl = fanotify_ioctl,
 443         .compat_ioctl   = fanotify_ioctl,
 444         .llseek         = noop_llseek,
 445 };
 446
 447 static void fanotify_free_mark(struct fsnotify_mark *fsn_mark)
 448 {
 449         kmem_cache_free(fanotify_mark_cache, fsn_mark);
 450 }
 451
 452 static int fanotify_find_path(int dfd, const char __user *filename,
 453                               struct path *path, unsigned int flags)
 454 {
 455         int ret;
 456
 457         pr_debug("%s: dfd=%d filename=%p flags=%x\n", __func__,
 458                  dfd, filename, flags);
 459
 460         if (filename == NULL) {
 461                 struct fd f = fdget(dfd);
 462
 463                 ret = -EBADF;
 464                 if (!f.file)
 465                         goto out;
 466
 467                 ret = -ENOTDIR;
 468                 if ((flags & FAN_MARK_ONLYDIR) &&
 469                     !(S_ISDIR(file_inode(f.file)->i_mode))) {
 470                         fdput(f);
 471                         goto out;
 472                 }
 473
 474                 *path = f.file->f_path;
 475                 path_get(path);
 476                 fdput(f);
 477         } else {
 478                 unsigned int lookup_flags = 0;
 479
 480                 if (!(flags & FAN_MARK_DONT_FOLLOW))
 481                         lookup_flags |= LOOKUP_FOLLOW;
 482                 if (flags & FAN_MARK_ONLYDIR)
 483                         lookup_flags |= LOOKUP_DIRECTORY;
 484
 485                 ret = user_path_at(dfd, filename, lookup_flags, path);
 486                 if (ret)
 487                         goto out;
 488         }
 489
 490         /* you can only watch an inode if you have read permissions on it */
 491         ret = inode_permission(path->dentry->d_inode, MAY_READ);
 492         if (ret)
 493                 path_put(path);
 494 out:
 495         return ret;
 496 }
 497
 498 static __u32 fanotify_mark_remove_from_mask(struct fsnotify_mark *fsn_mark,
 499                                             __u32 mask,
 500                                             unsigned int flags,
 501                                             int *destroy)
 502 {
 503         __u32 oldmask = 0;
 504
 505         spin_lock(&fsn_mark->lock);
 506         if (!(flags & FAN_MARK_IGNORED_MASK)) {
 507                 __u32 tmask = fsn_mark->mask & ~mask;
 508
 509                 if (flags & FAN_MARK_ONDIR)
 510                         tmask &= ~FAN_ONDIR;
 511
 512                 oldmask = fsn_mark->mask;
 513                 fsnotify_set_mark_mask_locked(fsn_mark, tmask);
 514         } else {
 515                 __u32 tmask = fsn_mark->ignored_mask & ~mask;
 516                 if (flags & FAN_MARK_ONDIR)
 517                         tmask &= ~FAN_ONDIR;
 518
 519                 fsnotify_set_mark_ignored_mask_locked(fsn_mark, tmask);
 520         }
 521         *destroy = !(fsn_mark->mask | fsn_mark->ignored_mask);
 522         spin_unlock(&fsn_mark->lock);
 523
 524         return mask & oldmask;
 525 }
 526
 527 static int fanotify_remove_vfsmount_mark(struct fsnotify_group *group,
 528                                          struct vfsmount *mnt, __u32 mask,
 529                                          unsigned int flags)
 530 {
 531         struct fsnotify_mark *fsn_mark = NULL;
 532         __u32 removed;
 533         int destroy_mark;
 534
 535         mutex_lock(&group->mark_mutex);
 536         fsn_mark = fsnotify_find_vfsmount_mark(group, mnt);
 537         if (!fsn_mark) {
 538                 mutex_unlock(&group->mark_mutex);
 539                 return -ENOENT;
 540         }
 541
 542         removed = fanotify_mark_remove_from_mask(fsn_mark, mask, flags,
 543                                                  &destroy_mark);
 544         if (destroy_mark)
 545                 fsnotify_detach_mark(fsn_mark);
 546         mutex_unlock(&group->mark_mutex);
 547         if (destroy_mark)
 548                 fsnotify_free_mark(fsn_mark);
 549
 550         fsnotify_put_mark(fsn_mark);
 551         if (removed & real_mount(mnt)->mnt_fsnotify_mask)
 552                 fsnotify_recalc_vfsmount_mask(mnt);
 553
 554         return 0;
 555 }
 556
 557 static int fanotify_remove_inode_mark(struct fsnotify_group *group,
 558                                       struct inode *inode, __u32 mask,
 559                                       unsigned int flags)
 560 {
 561         struct fsnotify_mark *fsn_mark = NULL;
 562         __u32 removed;
 563         int destroy_mark;
 564
 565         mutex_lock(&group->mark_mutex);
 566         fsn_mark = fsnotify_find_inode_mark(group, inode);
 567         if (!fsn_mark) {
 568                 mutex_unlock(&group->mark_mutex);
 569                 return -ENOENT;
 570         }
 571
 572         removed = fanotify_mark_remove_from_mask(fsn_mark, mask, flags,
 573                                                  &destroy_mark);
 574         if (destroy_mark)
 575                 fsnotify_detach_mark(fsn_mark);
 576         mutex_unlock(&group->mark_mutex);
 577         if (destroy_mark)
 578                 fsnotify_free_mark(fsn_mark);
 579
 580         /* matches the fsnotify_find_inode_mark() */
 581         fsnotify_put_mark(fsn_mark);
 582         if (removed & inode->i_fsnotify_mask)
 583                 fsnotify_recalc_inode_mask(inode);
 584
 585         return 0;
 586 }
 587
 588 static __u32 fanotify_mark_add_to_mask(struct fsnotify_mark *fsn_mark,
 589                                        __u32 mask,
 590                                        unsigned int flags)
 591 {
 592         __u32 oldmask = -1;
 593
 594         spin_lock(&fsn_mark->lock);
 595         if (!(flags & FAN_MARK_IGNORED_MASK)) {
 596                 __u32 tmask = fsn_mark->mask | mask;
 597
 598                 if (flags & FAN_MARK_ONDIR)
 599                         tmask |= FAN_ONDIR;
 600
 601                 oldmask = fsn_mark->mask;
 602                 fsnotify_set_mark_mask_locked(fsn_mark, tmask);
 603         } else {
 604                 __u32 tmask = fsn_mark->ignored_mask | mask;
 605                 if (flags & FAN_MARK_ONDIR)
 606                         tmask |= FAN_ONDIR;
 607
 608                 fsnotify_set_mark_ignored_mask_locked(fsn_mark, tmask);
 609                 if (flags & FAN_MARK_IGNORED_SURV_MODIFY)
 610                         fsn_mark->flags |= FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY;
 611         }
 612         spin_unlock(&fsn_mark->lock);
 613
 614         return mask & ~oldmask;
 615 }
 616
 617 static struct fsnotify_mark *fanotify_add_new_mark(struct fsnotify_group *group,
 618                                                    struct inode *inode,
 619                                                    struct vfsmount *mnt)
 620 {
 621         struct fsnotify_mark *mark;
 622         int ret;
 623
 624         if (atomic_read(&group->num_marks) > group->fanotify_data.max_marks)
 625                 return ERR_PTR(-ENOSPC);
 626
 627         mark = kmem_cache_alloc(fanotify_mark_cache, GFP_KERNEL);
 628         if (!mark)
 629                 return ERR_PTR(-ENOMEM);
 630
 631         fsnotify_init_mark(mark, fanotify_free_mark);
 632         ret = fsnotify_add_mark_locked(mark, group, inode, mnt, 0);
 633         if (ret) {
 634                 fsnotify_put_mark(mark);
 635                 return ERR_PTR(ret);
 636         }
 637
 638         return mark;
 639 }
 640
 641
 642 static int fanotify_add_vfsmount_mark(struct fsnotify_group *group,
 643                                       struct vfsmount *mnt, __u32 mask,
 644                                       unsigned int flags)
 645 {
 646         struct fsnotify_mark *fsn_mark;
 647         __u32 added;
 648
 649         mutex_lock(&group->mark_mutex);
 650         fsn_mark = fsnotify_find_vfsmount_mark(group, mnt);
 651         if (!fsn_mark) {
 652                 fsn_mark = fanotify_add_new_mark(group, NULL, mnt);
 653                 if (IS_ERR(fsn_mark)) {
 654                         mutex_unlock(&group->mark_mutex);
 655                         return PTR_ERR(fsn_mark);
 656                 }
 657         }
 658         added = fanotify_mark_add_to_mask(fsn_mark, mask, flags);
 659         mutex_unlock(&group->mark_mutex);
 660
 661         if (added & ~real_mount(mnt)->mnt_fsnotify_mask)
 662                 fsnotify_recalc_vfsmount_mask(mnt);
 663
 664         fsnotify_put_mark(fsn_mark);
 665         return 0;
 666 }
 667
 668 static int fanotify_add_inode_mark(struct fsnotify_group *group,
 669                                    struct inode *inode, __u32 mask,
 670                                    unsigned int flags)
 671 {
 672         struct fsnotify_mark *fsn_mark;
 673         __u32 added;
 674
 675         pr_debug("%s: group=%p inode=%p\n", __func__, group, inode);
 676
 677         /*
 678          * If some other task has this inode open for write we should not add
 679          * an ignored mark, unless that ignored mark is supposed to survive
 680          * modification changes anyway.
 681          */
 682         if ((flags & FAN_MARK_IGNORED_MASK) &&
 683             !(flags & FAN_MARK_IGNORED_SURV_MODIFY) &&
 684             (atomic_read(&inode->i_writecount) > 0))
 685                 return 0;
 686
 687         mutex_lock(&group->mark_mutex);
 688         fsn_mark = fsnotify_find_inode_mark(group, inode);
 689         if (!fsn_mark) {
 690                 fsn_mark = fanotify_add_new_mark(group, inode, NULL);
 691                 if (IS_ERR(fsn_mark)) {
 692                         mutex_unlock(&group->mark_mutex);
 693                         return PTR_ERR(fsn_mark);
 694                 }
 695         }
 696         added = fanotify_mark_add_to_mask(fsn_mark, mask, flags);
 697         mutex_unlock(&group->mark_mutex);
 698
 699         if (added & ~inode->i_fsnotify_mask)
 700                 fsnotify_recalc_inode_mask(inode);
 701
 702         fsnotify_put_mark(fsn_mark);
 703         return 0;
 704 }
 705
 706 /* fanotify syscalls */
 707 SYSCALL_DEFINE2(fanotify_init, unsigned int, flags, unsigned int, event_f_flags)
 708 {
 709         struct fsnotify_group *group;
 710         int f_flags, fd;
 711         struct user_struct *user;
 712         struct fanotify_event_info *oevent;
 713
 714         pr_debug("%s: flags=%d event_f_flags=%d\n",
 715                 __func__, flags, event_f_flags);
 716
 717         if (!capable(CAP_SYS_ADMIN))
 718                 return -EPERM;
 719
 720         if (flags & ~FAN_ALL_INIT_FLAGS)
 721                 return -EINVAL;
 722
 723         if (event_f_flags & ~FANOTIFY_INIT_ALL_EVENT_F_BITS)
 724                 return -EINVAL;
 725
 726         switch (event_f_flags & O_ACCMODE) {
 727         case O_RDONLY:
 728         case O_RDWR:
 729         case O_WRONLY:
 730                 break;
 731         default:
 732                 return -EINVAL;
 733         }
 734
 735         user = get_current_user();
 736         if (atomic_read(&user->fanotify_listeners) > FANOTIFY_DEFAULT_MAX_LISTENERS) {
 737                 free_uid(user);
 738                 return -EMFILE;
 739         }
 740
 741         f_flags = O_RDWR | FMODE_NONOTIFY;
 742         if (flags & FAN_CLOEXEC)
 743                 f_flags |= O_CLOEXEC;
 744         if (flags & FAN_NONBLOCK)
 745                 f_flags |= O_NONBLOCK;
 746
 747         /* fsnotify_alloc_group takes a ref.  Dropped in fanotify_release */
 748         group = fsnotify_alloc_group(&fanotify_fsnotify_ops);
 749         if (IS_ERR(group)) {
 750                 free_uid(user);
 751                 return PTR_ERR(group);
 752         }
 753
 754         group->fanotify_data.user = user;
 755         atomic_inc(&user->fanotify_listeners);
 756
 757         oevent = fanotify_alloc_event(NULL, FS_Q_OVERFLOW, NULL);
 758         if (unlikely(!oevent)) {
 759                 fd = -ENOMEM;
 760                 goto out_destroy_group;
 761         }
 762         group->overflow_event = &oevent->fse;
 763
 764         if (force_o_largefile())
 765                 event_f_flags |= O_LARGEFILE;
 766         group->fanotify_data.f_flags = event_f_flags;
 767 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 768         spin_lock_init(&group->fanotify_data.access_lock);
 769         init_waitqueue_head(&group->fanotify_data.access_waitq);
 770         INIT_LIST_HEAD(&group->fanotify_data.access_list);
 771 #endif
 772         switch (flags & FAN_ALL_CLASS_BITS) {
 773         case FAN_CLASS_NOTIF:
 774                 group->priority = FS_PRIO_0;
 775                 break;
 776         case FAN_CLASS_CONTENT:
 777                 group->priority = FS_PRIO_1;
 778                 break;
 779         case FAN_CLASS_PRE_CONTENT:
 780                 group->priority = FS_PRIO_2;
 781                 break;
 782         default:
 783                 fd = -EINVAL;
 784                 goto out_destroy_group;
 785         }
 786
 787         if (flags & FAN_UNLIMITED_QUEUE) {
 788                 fd = -EPERM;
 789                 if (!capable(CAP_SYS_ADMIN))
 790                         goto out_destroy_group;
 791                 group->max_events = UINT_MAX;
 792         } else {
 793                 group->max_events = FANOTIFY_DEFAULT_MAX_EVENTS;
 794         }
 795
 796         if (flags & FAN_UNLIMITED_MARKS) {
 797                 fd = -EPERM;
 798                 if (!capable(CAP_SYS_ADMIN))
 799                         goto out_destroy_group;
 800                 group->fanotify_data.max_marks = UINT_MAX;
 801         } else {
 802                 group->fanotify_data.max_marks = FANOTIFY_DEFAULT_MAX_MARKS;
 803         }
 804
 805         fd = anon_inode_getfd("[fanotify]", &fanotify_fops, group, f_flags);
 806         if (fd < 0)
 807                 goto out_destroy_group;
 808
 809         return fd;
 810
 811 out_destroy_group:
 812         fsnotify_destroy_group(group);
 813         return fd;
 814 }
 815
 816 SYSCALL_DEFINE5(fanotify_mark, int, fanotify_fd, unsigned int, flags,
 817                               __u64, mask, int, dfd,
 818                               const char  __user *, pathname)
 819 {
 820         struct inode *inode = NULL;
 821         struct vfsmount *mnt = NULL;
 822         struct fsnotify_group *group;
 823         struct fd f;
 824         struct path path;
 825         int ret;
 826
 827         pr_debug("%s: fanotify_fd=%d flags=%x dfd=%d pathname=%p mask=%llx\n",
 828                  __func__, fanotify_fd, flags, dfd, pathname, mask);
 829
 830         /* we only use the lower 32 bits as of right now. */
 831         if (mask & ((__u64)0xffffffff << 32))
 832                 return -EINVAL;
 833
 834         if (flags & ~FAN_ALL_MARK_FLAGS)
 835                 return -EINVAL;
 836         switch (flags & (FAN_MARK_ADD | FAN_MARK_REMOVE | FAN_MARK_FLUSH)) {
 837         case FAN_MARK_ADD:              /* fallthrough */
 838         case FAN_MARK_REMOVE:
 839                 if (!mask)
 840                         return -EINVAL;
 841                 break;
 842         case FAN_MARK_FLUSH:
 843                 if (flags & ~(FAN_MARK_MOUNT | FAN_MARK_FLUSH))
 844                         return -EINVAL;
 845                 break;
 846         default:
 847                 return -EINVAL;
 848         }
 849
 850         if (mask & FAN_ONDIR) {
 851                 flags |= FAN_MARK_ONDIR;
 852                 mask &= ~FAN_ONDIR;
 853         }
 854
 855 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 856         if (mask & ~(FAN_ALL_EVENTS | FAN_ALL_PERM_EVENTS | FAN_EVENT_ON_CHILD))
 857 #else
 858         if (mask & ~(FAN_ALL_EVENTS | FAN_EVENT_ON_CHILD))
 859 #endif
 860                 return -EINVAL;
 861
 862         f = fdget(fanotify_fd);
 863         if (unlikely(!f.file))
 864                 return -EBADF;
 865
 866         /* verify that this is indeed an fanotify instance */
 867         ret = -EINVAL;
 868         if (unlikely(f.file->f_op != &fanotify_fops))
 869                 goto fput_and_out;
 870         group = f.file->private_data;
 871
 872         /*
 873          * group->priority == FS_PRIO_0 == FAN_CLASS_NOTIF.  These are not
 874          * allowed to set permissions events.
 875          */
 876         ret = -EINVAL;
 877         if (mask & FAN_ALL_PERM_EVENTS &&
 878             group->priority == FS_PRIO_0)
 879                 goto fput_and_out;
 880
 881         if (flags & FAN_MARK_FLUSH) {
 882                 ret = 0;
 883                 if (flags & FAN_MARK_MOUNT)
 884                         fsnotify_clear_vfsmount_marks_by_group(group);
 885                 else
 886                         fsnotify_clear_inode_marks_by_group(group);
 887                 goto fput_and_out;
 888         }
 889
 890         ret = fanotify_find_path(dfd, pathname, &path, flags);
 891         if (ret)
 892                 goto fput_and_out;
 893
 894         /* inode held in place by reference to path; group by fget on fd */
 895         if (!(flags & FAN_MARK_MOUNT))
 896                 inode = path.dentry->d_inode;
 897         else
 898                 mnt = path.mnt;
 899
 900         /* create/update an inode mark */
 901         switch (flags & (FAN_MARK_ADD | FAN_MARK_REMOVE)) {
 902         case FAN_MARK_ADD:
 903                 if (flags & FAN_MARK_MOUNT)
 904                         ret = fanotify_add_vfsmount_mark(group, mnt, mask, flags);
 905                 else
 906                         ret = fanotify_add_inode_mark(group, inode, mask, flags);
 907                 break;
 908         case FAN_MARK_REMOVE:
 909                 if (flags & FAN_MARK_MOUNT)
 910                         ret = fanotify_remove_vfsmount_mark(group, mnt, mask, flags);
 911                 else
 912                         ret = fanotify_remove_inode_mark(group, inode, mask, flags);
 913                 break;
 914         default:
 915                 ret = -EINVAL;
 916         }
 917
 918         path_put(&path);
 919 fput_and_out:
 920         fdput(f);
 921         return ret;
 922 }
 923
 924 #ifdef CONFIG_COMPAT
 925 COMPAT_SYSCALL_DEFINE6(fanotify_mark,
 926                                 int, fanotify_fd, unsigned int, flags,
 927                                 __u32, mask0, __u32, mask1, int, dfd,
 928                                 const char  __user *, pathname)
 929 {
 930         return sys_fanotify_mark(fanotify_fd, flags,
 931 #ifdef __BIG_ENDIAN
 932                                 ((__u64)mask0 << 32) | mask1,
 933 #else
 934                                 ((__u64)mask1 << 32) | mask0,
 935 #endif
 936                                  dfd, pathname);
 937 }
 938 #endif
 939
 940 /*
 941  * fanotify_user_setup - Our initialization function.  Note that we cannot return
 942  * error because we have compiled-in VFS hooks.  So an (unlikely) failure here
 943  * must result in panic().
 944  */
 945 static int __init fanotify_user_setup(void)
 946 {
 947         fanotify_mark_cache = KMEM_CACHE(fsnotify_mark, SLAB_PANIC);
 948         fanotify_event_cachep = KMEM_CACHE(fanotify_event_info, SLAB_PANIC);
 949 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 950         fanotify_perm_event_cachep = KMEM_CACHE(fanotify_perm_event_info,
 951                                                 SLAB_PANIC);
 952 #endif
 953
 954         return 0;
 955 }
 956 device_initcall(fanotify_user_setup);